As with all uses of “just” I meant “With a huge amount of technical and political effort”.
It’s true that Deep Learning is extremely opaque, but that just means it’s probably not the right tool for this job. There are plenty of interpretable models between a simple linear regression and full on deep learning.
In the machine learning community I think we’ve become over eager to sacrifice any ability to audit a system in exchange for another point of accuracy.
The example of higher arrest rates within certain communities is an interesting example of exactly the sorts of things we could be alleviating rather than exacerbating with automated job-seeker evaluation. There’s been plenty of sociological work on the impact of wealth and race on criminal convictions. If we decide that, say, race shouldn’t be a factor in convictions, then we can calculate the probability a poor black man would not have been arrested had he been a poor white man. Then we can say something like “ok, 60% of the time that arrest is a huge red flag, 20% of the time it’s something to look out for, 20% of the time we can dismiss it as institutional racism.” Then it counts against any individual, but less given their circumstances. And if the rest of the resume is strong it can be counteracted. And even if the rest of the resume is so/so, a good reinforcement algorithm would still hire the person in rare instances to figure out if there’s something it’s missing here. Which is a huge difference from “criminal conviction = waste basket” that current algorithms and human HR managers do.
Algorithms can be improved, subtle biases can be identified. If things are technically as bad as people’s experiences and research suggest it’s a shame these companies don’t have the basic decency to do the right thing and approach the problem delicately, professionally and humanely.
Chicago here… back in the “dizzay” when I worked as a carp the worst possible “worker” was the 3rd/4th generation Irish kid who would spend his days toiling away at telling us how hard his grandfather had it. At least his grandfather knew what a hammer looked like.
No, they really won’t. Not as long as the people with the money hold all the bargaining power.
For my entire career in my (very “good”) field, 90% of the job listings require 10 years of experience in things that have only existed for 6 years, a doctorate in a related field, and three trips to Mars. For mid-level retail wages or high-ish non-management fast food wages.
They don’t want to hire. Not outside of H1Bs anyway. “We can’t find qualified staff in the U.S.” Riiiiight. Back when I was involved in part of the hiring process, 75% of the people I interviewed were overwhelmingly overqualified. They interviewed because we pay living wage.
These folks aren’t failing for excluding awesome potential employees. They just make the ones they already have do 2-3x the work.
P.S. I work for a really great company and having a fabulous time. For the first time in my career.
Yeah I have kinda boggled at some of the salaries stated for given work statement… I was like wait man counting for inflation I made better than that starting out as an admin only having to wrangle only CC:Mail and I am sure they were happy cause I was cheap at the time. I know none of my ex coworkers would do Exchange + Active Directory + SQL + automated patch management + backups + etc for less than 70k but I have seen places that want to offer 45k for that. Just like when the app group I was last with had some job requirements out for like experienced support folk but were offering entry level wages and were clueless why nobody was applying as they needed to have actual employees rather than contractors for the position.
For those companies, these algorithms are working exactly as intended. But a company actually seeking to hire quality employees will put themselves at a competitive disadvantage by using formulae that exclude anyone over 35, or anyone with a black sounding name, or anyone not lying about having ten years’ experience with a six-year-old product.
Still. Hasn’t hurt them yet. If you have money and want good, under-appreciated employees, they’re not hard to find … at all. There’s no competitive disadvantage to not hiring awesome potential employees if no one else is going to hire them either.
Haha. HP, Yahoo, and the other barely-theres of the IT world aren’t floundering because of hiring decisions (at least not front line hiring decisions … management, maybe yes). I don’t know how to convey this tone except by writing it out fully. I genuinely laughed. Not a disagreement laugh. Consider it mutual schadenfreude for the woes of HP and Yahoo if you want. Can we throw Dell on this list too?
They make pretty good server hardware and monitoring software but service manager can die in fucking fire… my ex-employer is about to buy/merge with the enterprise services and get Meg Whitman in the deal so while unemployment sucks at least I am not working for her.
Yep. I edited my post about the same time you were posting.
To bring it back on topic, it doesn’t matter in the slightest if they use an algorithm to discriminate against people or use HR staff who have an aversion to calling people with non-white sounding names. The algorithm makes it faster (not cheaper probably) to ignore awesome candidates for discriminatory reasons. The effect is the same.
Massive amounts of awesome potential hires who can’t get a break and theoretical near infinity of opportunities lost to the companies who won’t hire them.
My first year out of high school involved applying for several dozen jobs every week (anything that I had a chance at; pretty much all minimum wage and brainless), getting interviews for two or three of them, and not getting hired. Every week, for a year.
When I finally did get a job, the annual salary was AU$13,000 (equivalent to about US$8K at the time). In the 1990’s.